ZEMCH 2015 - International Conference Proceedings | Page 85

Figure 3: Global energy performance related cost for various wall envelope technologies
To quantify uncertainty, i.e. uncertainty analysis (UA), amount of outcome values, first standard
deviation (SD) was utilized. SD is a proper measure to describe data dispersion around the mean
and has the same unit of data. In figure.4, mean value for global cost obtained from 100 times of
Monte Carlo simulations is shown. Moreover, SD of all output data of 22 cases respect to their
mean value is presented which varies within 30 to 36 €/m2 and its magnitude is about 5% of the
mean.
However SD is a practical statistical indicator for data distribution, it doesn’t consider the importance of variation respect to the actual value and in this study acts mostly as an indicator for
comparison of wall alternatives. Hence, for a more overall perception of results, a probabilistic
approach was considered. In figure.5, box whisker plot graph visualize a proper overview of uncertainty distribution in obtained outcome. The identical range of output as well as equal length
of interquartile for all cases indicate that differences in wall technology, which ends in various
global costs, doesn’t influence tendency of output uncertainty. Distribution of data in all cases fits
normal distribution due to median and mean value located at the same level and no skewness is
observed in none of them. Therefore, assuming that output distribution is a normal one, it could
be concluded that 68% of data (± SD) is located within ±5% on either side of the mean and in the
same way, 95% of data (± 2SD) within ±10.
Additionally, apart from total uncertainty quantification, it is an added value to distinguish individual input importance in output uncertainty and figure out which input parameters are more
dominant than others in varying outcome. Hence in the second step, a sensitivity analysis (SA)
was performed to study how output variation could be attributed to variation in individual input
parameters. For this purpose the regression analysis was utilized since it shows more quantitative
measures of sensitivity.

Uncertainty effects of input data on cost optimal NZEB performance analysis

83